On the Lower of the Set-Valued
Semicontinuity Metric Projection
BRLNO
BROSOWSKI
AND RUDOLF
DEDICATED
TO
PROFtSSOR 01:
WEGMANN
I.
1.
J.
SCHOCNBERG
701-H
HIS
THE
OCCASION
INTRODUCTION
Let X be a real normed linear space and each elementf’in X we associate the set P,(f)
ON
BIRIHDAY
:= {P” E v : ,i,f’--
I/’ a proximinal
I’(, 1~= gy: lL 1,I’--
subset of A’. To I’$,
which is called the set of best approximations for f by elements of V. Thus we obtain a set-valued mapping P,, , which carries the normed linear space X into the set of the closed nonvoid subsets of V. This set-valued mapping is called the metric prqjection associatedwith V. For set-valued mappings concepts of continuity are dehned as follows (cf. Hahn [5]): DEFINITION
continuous
1. (a) The (USC) if the set
metric
projection
{f’c X : P&f)
P,(f)n
84 Copyright All rights
0 1973 by Academic Press, Inc. of reproduction in any form reserved.
upper
semi-
n K + 17:
is closed whenever K is a closed subset of V. (b) The metric projection is called lower {fEX:
Pv is called
U -i
semicontinuous a}
(kc) if the set
CONTINUITY
OF THE METRIC PROJECTION
85
is open whenever U is an open subsetof V. (The topology on V is understood to be induced by the norm-topology of X). The metric projection Pv is USCor Isc only for restricted classesof subsetsI/. Singer [8], for example, has proved that the metric projection associatedwith an approximatively compact subset I/ of a normed linear spaceis USC.Hence, in particular, Pv is USCwhenever C’ is a linear subspaceof finite dimension. But even if V is a linear subspaceof finite dimension, P,, may fail to be IX, as Blatter, Morris, and Wulbert [2] have shown. In this paper, we first prove a genera1criterion which is sufficient for the lower semicontinuity of the metric projection associated with certain subspacesof a normed linear space. As a consequenceof this criterion, we obtain a result of Brosowski rt al. [4]. Further we apply our criterion to derive ;I sufficient condition for the lower semicontinuity of the metric projection associated with certain linear subspacesof C,,(T, X), where T is a locally compact Hausdorff-space, X is a strictly convex normed linear space, and C&7: X) is the set of all continuous functions .f: T --+ X which vanish at infinity, provided with the norm I;.f’l\ :- maxtSTli,f(r)l;x . We shall show that in C,,(T, X) the criterion thus obtained is also necessary for the lower semicontinuity of PV. This generalizesthe results of Blatter [I], Blatter, Morris, and Wulbert [2], and Brosowski et al. [3]. Furthermore, we apply our genera1 sufficient criterion to prove the sufficiency of a criterion stated by Lazar, Wulbert, and Morris [6] for the lower semicontinuity of the metric projection associated with fnitedimensional linear subspacesof L,( r, S, p), where (T, R, CL)is a o-finite measure space.
2. A SUFFICIENT CONDITION
FOR THE LOWER SEMICONTINUITY
OF Py
We first state some necessary definitions, For X a normed linear space, we define Sx := {xEX:
!ixII -<, 1)
and 6, := Ep(SX) where Ep(A) denotes the set of extreme points of a set A. ForfE X, we set z; := {x’ ESX’ : x’(f)
= :Ifll)
G”,:= {x’ EBJy : x’(f)
= llflj}.
and
86
BROSOWSKI
AND
WEGMANX
As is well known, c?, == 2, n A,’ . We use the terms u-topology and o,,,topology to denote the restrictions of the weak topology 0(X’, X) on the sets S,, and AX’ , respectively. For V a proximinal linear subspaceof X andf‘in .Y with 0 t’ p,.(,f‘). we define the set
The subscript V will be omitted if it is understood from the context. Finally, we define E(,fi V) :=- &f (. f
I’ / .
Now we prove the following. LEMMA 2. Let A be u subset sf’ t:.,’
and ,f’
WI
elemetlt
itI .Y. Tlwn
tlw
inequulil~l~
sup s’(f) \‘i 1 holds ~fiwl
Proof:
onlv_ if‘theve exists u
(T(
X’.
X)-open
f c~mrer
subset
C:
0J.Y
surli
fhut
Let A be a subsetof 6 Y’ Whenever
then there is an E .:’ 0 so that s . ~:,/‘, set u :=- {.u’ E X’ : s’(f) we have A n lJ :
E. For the 0(X”. .Y)-open convex .f’
;? , and hence A n (L’ 17d,!,)
t:. Thus we obtain
To prove the inverse implication. we assumethat C’ is a u(X’, X)-open convex subset of X’ such that
CONTINUITY
OF
THE
METRIC
PROJECTION
87
The Krein-Milman theorem yields Z, : con gI. Since Ii is convex, and contains 6, , there follows U 3 con L> . We show that U3 L?~ In fact, suppose that there is an element -Y,,’ in -‘f which is not in U, Then x0’ is in the boundary of U, and, since U is open and convex, there exists an element ‘(g). Define H : = (x’ E X’ : x’(g) y= x,‘(g)]. Since U is open, H n I/ =y iY. Using 2, C l? we obtain
whence H n Zf is an extremal subset of 2, . Therefore H n Z; contains an extreme point x1’ of &. In view of Lf C U we have x1’ E H n iJ. which contradicts H n U = 4:I. Thus we have 7J 3 Zf and finally, since .Sx-,\U is compact,
This completes the proof. We are now ready to prove the following. THEOREM 3. Let V be a proxitGm1 space X with the properties:
(I ) ,for each ,f E X, dim P,(f) (2)
linear subspace of a normed linear
< CO,
the metric projection P, is use.
Whenewr ,for each elementf in X with 0 E P,(f) cotll7e.x subset I/ of X’ suci~that
there exists a 0(X”, X)-open
then P, is Isc.
Proor We assumethat there exists a point f in X so that P,, is not lsc in f. Then there exists a sequence{f,,] of elements,f, in X, an element z’,, in PY(f 1 and a neighborhood U0of c0such that { f,J converges to f and
for each n. The set P,(f) is convex and. since P, is USC,consistsof more than one point. We may assumewithout loss of generality that u,, is a relative interior point of P,(f). Replacing f by f - t’,, if necessary, we may assume that z’,~= 0.
88
BROSOWSKI
AND
WtGUANX
Then it follows that
For each g E X the set PV(g) is closed. bounded and finite-dimensional. and hence compact. Since Pr, is USC.the set P‘,(f) u iJ pv-(.fn) ,,BY is compact (cf. Michael [7]). Consequently there is a subsequenceof {fit; (which we denote again by (fr,l), so that there exist elements V, in P,(frt) in such a way that the sequence{c,) converges to an element 1:’ in Y. Then z.’E P,(f), and furthermore 1” ? 0 since PJf;,) n b’,, 7 for all ME N. Following Brosowski et al. [4], we can now construct a sequence{u,J of elementsu,! in X with the following properties: (a) the sequence{u,) converges to 1”: (b) for each n t N there holds f’ - II,, /! (c)
E(fi
C):
there exists an integer M,,E N so that
I:.f -- uII I- 1“ 1,_” E(,f: V)
for all II
II,,
Such a sequencemay be defined explicitly by 11,: =
.f - (E(f; V)/E(f, : L’))(,f,
!.,,I.
Obviously this sequencehas properties (a) and (b). To show that {u,(; also sharesproperty (c), we note that 1i.f- u, --I-r’ !~ _ E(,$ V), since .f - Ll,, .. 1.’-= (E(fi w-w;,
; VW;, -- (l’,, ~ (E(,f,, : V),‘E(fi
V)) 1.‘)).
(I)
If there is some subsequenceof {II,,/ (again denoted by 1~1,~)) such that IIf - u,, -!- I’’ :/ : E(f; V). then, by (I ). the element
is in P,(f,t). and the sequence{O,,] converges to the zero element in V. This is impossible since P,,(f,J n U, = D for each n. Hence (c) is proved. For n 3‘. fr,, and x’ E J,--l,,, !,’ we have E(f; V) < IIf- u,, t- 1.’‘, = x’(f - u,L I- r’) = x’(.f --- u,,) i E(f; V) -~ X’(1”).
.Y’(l.‘)
CONTlNUlTY
OF
l-HE
METRIC
PROJECTION
8’)
which yields XI(&) > 0. Hence we obtain
On the other hand, by hypothesis, subset U of X’ so that
there exists
a 4X’, X)-open
convex
and, by Lemma 2,
Hence there exists a real number E “2, 0 such that
for each x’ E CX,\CJ. The sequence {u,,) converges to L”. Therefore ,: II,, - u’ 1; < E for sufficiently large n, and consequently x’(.f - 71,; ,- r’) < E(f
we have
V)
for all x’ in dX,\U, in particular for x’ in G,J\%, . Thus we obtain for sufficiently large n the inclusion 91, 3 &j-,,n+V, which contradicts (2). The theorem is thus proved. A consequence of Theorem 3 is the following result of Brosowski et 01. (41. linear subspaceoj’ a normed lirlear THEOREM 4. Let V be a prosimiml space X with the properties (I), (2) of’ Theorem 3 arid the Jdlowing additiotzal
property: The sef L ,ynis o-dosed. Wheneverfor each elementf in .X’with 0 E P,(f) of Sx, such that
(3)
there exists a o-open subsetA
then the metric projection Py is 1s~. Proof.
640/8/1-S
In view of property (3), the set GX, is a-compact. Hence we obtain
90
BKOSOWSKI
,\\I)
WtGMAhuh
and, using the hypothesis.
Thus, by Lemma 2, there is a a(X’, X)-open convex subset U of X’ such that Ylf3 lJn d,,’ 3 6, , and Theorem 3 yields the lower semicontinuity of P,.
3. APPLICATIONS
Ih
SOME
SPECIAL
SPACES
Let T be a locally compact HausdorH‘ space, and X a strictly convex real normed linear space. We denote by C,,( T. Y) the set of continuous functions ,f: T + X vanishing at infinity, that is. a continuous functionj‘is in C,,(T, X) if and only if. for each E ‘, 0, the set {te7‘:
f(t),
E:
is compact. If addition and multiplication with scalarsare defined for elements in C,(T, X) in the same way as for vector-valued functions, and the norm : f‘i; :
Ifa;r
f(f)
,t
is introduced. then C,( T, X) is a normed linear space.Whenever it is necessar) to distinguish between the norms in C,,(T. X) and X we denote the latter by Ii . 11x. For V a proximinal subspaceof C,,CT. X). f’an element in C,,(T, X) with 0 t P,(f), and I‘,, an element in P,,(,f). we define
and ;I1,-,.,, : : ;t E T: J(t)
1’(,(f),.\ -7 f‘
1’0!/.
The subscript V will be omitted if it is understood from the context. Now we prove the following. LEMMA 5. Let V be a proximinal litlear subspaceof C,,(7: A’), where X is strictly COIZWX.Then
./ix each eletuentf in C,(T, X) with 0 it1 P,-(f).
CONTINUll-Y
OF THE WETRIC PROJECTION
91
Proof. Let t be in nceP+) A[,,_(_ Since 0 E PV(f), for each I’ in P,(f) the element 4~’is also in P,(f). Hence Iif
- ;r(r)l, = f;lf(r),‘ ,- kiif(t) - r(t)l.
In view of the strict convexity of A’, there is a positive real number p such that
Using jif(t)li = /if(t) - r(t)!], we obtain p = 1 and finally l.(t) - 0, whence rtlv,. Now we give a sufficient criterion for the lower semicontinuity of P,, in G,(T Xl. THEOREM 6. Let V be a proximinal linear subspace of the space Z :-= C,,(T, X), where T is a locally compact Hausdorf space, and X is a strictly conrex space. Assume thaf the ,following requirementsboll:
(I) (2)
for
each f in Z, dim P,.(f)
-:: E;
the metric projection P, is use.
Whenecerfor eachf in Z ti,ith 0 E P,.(f) the set Nf is open then the metric projecrion is kc. Proof. The set Nf is defined to be the intersection of the closed sets it t T: z:(t) = O), 2:E P,(f), hence it is closed. Since Nf is also open by hypothesis, the function g defined by g(l) . _ t f(t) . (0
for for
f in N, , t in T\,N, ,
is in C,(r, X). The set
v:== jz’rZ’:z’(g)
> oj
is a a(Z’, Z)-open and convex subset of Z’. The functionals in 6=, are generalized evaluation functionals L,,,, , that is, there exist elementsx’ E 8,~ and t E T such that L,,,,(h) = x’@(t))
for
12E C,(T, X).
By definition of U, for each functional L,P,~E bz, n U, the equality L,,,,(r) = X’(L’(T)) = x’(0) = 0
holds for all I’ in P, (,f), whence we we obtain the inclusion
conclude
111, 3 C z’ n
Thus the requirements of Theorem 3 are fulfilled. continuity of P,, follows.
Lj.
Using Lemma 5.
whence the lower
semi-
Renm-ks. Special cases of Theorem 6 have appeared in the literature. For T compact and X the real axis, the theorem was proved by Blatter, Morris, and Wulbert [2]. For T locally compact and X a pre-Hilbert space. the result was obtained by Brosowski rt u/. [4]. Now let A’ be the space L,( T, R, p). where (T, JI. p) is a u-finite measure space. The dual space A” is identical to f.,( T. .G,p). For ,I’ a real-valued function defined on T, we set
supp(f) : :=~{r E T: f(t) Z(f)
:
{t E T: J(f)
S(f) :
(IE 7‘: f(t)
Oi, 0; sup i f(s), TFT
These setsare defined only up to setsof zero measure. In addition. we define for each linear subspaceV of X the orthogonal space V- ::
{x’ E X’ : x’(r) -= 0 for each 1’E VI
Lazar. Wulbert, and Morris [6] proved the following criterion THEOREM 7. Let (T, R, ,LL)be a crjkite t~leasurespace, ad let V he au n-dinwtwional liueur subspaceqf L,( T. St.p). The metric pmjectiolz P, is 1.w fund only if there doesnot e.xist a17x’ iI1 1,’ .Y’ I.m0. aid a L’iti Vfor which
(I)
S(x’) is pure[~~utomic, amI contnim 131/llo.st II -
I
atom,
(2) Z(0) co/lta;t?s S(x’), (3) supp (0) is not the union of u,fi~~iteJtimily of atom. We give a new proof for the sufficiency of this criterion by showing that the condition of Lazar, Wulbert, and Morris implies the condition of Theorem 3. From this, it follows in the case X -= L,(T, H, p) that the condition of Theorem 3 is also necessaryfor the lower semicontinuity of Pv . Proof of the suJiciency of the criterion in Theorem 7. We supposethat the condition of Theorem 7 holds but that there exists an element f E L, with 0 E pp(f) (without loss of generality we may even assumethat 0 is a relative
interior point of PI (f)) such that 91, does not contain 6 n 6 y’ whenever I: is a CT(X.‘.X)-open convex subset of A” with I/’ n 6 X’ 3 ~5~ First ue show that there exists an element i: in P,.(f) such that supp(?) is not the union of a finite family of atoms. To prove this. we suppose that for each I‘ in f’,;(f), the support supp(c) is a union of a finite number of atoms. Then there exist atoms A, ..... Ah such that supp(r.) C A := A, u ... u As for each ~1in P,.(f). For every r E P,.(f) and every x’ t LJ , we have x’(r) = 0. Slince the functionals x’ EC! (interpreted as functions in L,) may be chosen outside supp (,f’) arbitrarily retaining only the requirement I .v’(t) 1 = I, there must be supp(c) C supp(f) for all I‘ E f,,(f). Therefore one may assume A C supp(f). Corresponding to the set I+’ := ix’ E ~-5 ,y’ : x’(t) y= sign(,f(‘)) for t t A: there exists a 0(X’. X)-open convex subset U of X’ so that W --- Ci n d %-’, nameI!. e.g. Cl :=-~ :.r’ t X’ : 1’ .r’(t)(x,,$r) ‘T
. f(t)) dp : 0 for 1’ = I...., N’ \
where x..,&,denotes the characteristic function of the set A, . In addition. we have
Since such a relation was excluded by the choice of,l; we have proved our assertion that there exists some element 6 in Pr (,f) such that supp(.?) is not the union of a finite number of atoms. Now let Y’ :- {x’ ES-~, : x’(r) = 0 for all 13E V, x’(t) 7 sign(.f(t)) for t E supp(f)l, then Y’ is convex and n-compact. By the well-known theorem of characterization of best approximations, there exists a functional x’ EL’f such that .u’(z.) -~- 0 for all 1’ in V. Since this x’ is in Y’, the set Y’ is nonvoid. Hence there exists some extreme point J*’ of Y’. By construction, we have / ~“(t)~ m=;I for I c supp(f). Since supp(d) C supp(f) it follows that Z(G) 3 S(J*‘). Now we show that S(J,‘) is purely atomic and consists of at most II - 1 atoms. Let G, ri .. .._ P’ be a basisfor V with G = F.
94
BROSOWSKI AND WCC\fANN
First we exclude that S( I”) contains a nonatomic part. In fact. let B II S( J’) be a nonatomic subsetof S($) with p(B) : 0. Then there exists E :- 0 so that B, :- {r i- B : ,-v’(t)
I
E;
has p(B,) ;_’ 0. Then there exists a function :’ E .Sx. , :’ 0, such that supp(z’) C B, and JR1:‘(t) l.‘(t) l/p 0 for i I..... 17. l-or either sign. y’ It EZ’is in Y’. This contradicts the fact that J.’ is an extreme point of I;‘. Now let us supposeS(J,‘)T) B =~ B, u .‘. u B,, , where B,. are atoms. It follows that E := I - ess-sup,,R: $(t)i . 0. Let P,’ be the value of L.~on the atom B, . The system of equalities
has a nonzero solution 01~(‘,....\,,I’ with sign the function I’, defined by
lrI:) ~
\ y’(1) -:- tklL.O z’(t) :-~ , J,‘(r)
1 for all for for
i'.
Hence for either
t t B, t$B’
is in Y’. This is impossiblesinceJ.’ is an extreme point of k”. Therefore S( y’) contains at most II ~ 1 atoms. So far, we have constructed elements d E I’ and J“ ‘- Y’ 81 c’ which fulfil the requirements (I), (2), and (3) of Theorem 7. But the condition of Theorem 7 states that such elementsdo not exist. Hence our assumption is not correct. Thus we have proved, that the condition of Lazar. Wulbert. and Morris implies the condition of Theorem 3. Since the latter is sufficient for the lower semicontinuity of P,, the sufficiency part of Theorem 7 is proved.
4. THE NECESSITY OF THE CRITERION IN THE SPACE C,,(T, X) In this paragraph, we show that the sufficient condition of Theorem 6 is also necessaryif the spaceunder consideration is C,( T. X). where X is strictly convex. First we prove the following lemma. Lef V be a linear subspaceof‘C,,(T, X), and let,f undg be elements in C(,(T, X) M’ith LEMMA
8.
(a) i1.f ~11 (1g F; (b) 0 E P,(f) (c)
and 0 E PtJ(g-):
there is a neighbotrrhoodU of Mf such tlmtf (t)
g(t) for t E c’.
Proc!f:
Given an element I’ =T-0 in P,,(R). let h be the positive number
By virtue of hypothe& t in C
(b). the element r, : -- AL.is also in PI (g). We have for
I J‘(r) - r.,(t)
-= j g(t)
/‘,(I)’
’ 1 K = l,,f ‘.
and for t E T’ I f(f)
~-- r.l(t)
1 .f(l)
I r,(t)
E*
(’ .f’
~
E*)
--: i!.f“,
Hence the element 1.r is a best approximation for .f; and I’ = (I/h) Y, is in span P,.(f). We are now in position to prove the main result of this paragraph. ~‘HCOREM 9. Let T be a local/v compact cot~w.x norn7ed litlear space, and let C’ be a C,,( T. X) such that dim P,(f) < E ,fkv all f in prqjeclion P, is 1.~. then,,for each,f in C,,( T, X)
Hausdor# space, X a strictly proxirninal liuear subspace of C,( T, X). Whenever the metric \c,ith 0 in P,(f), the set
Ploc$ We suppose the theorem is false, that is, there exists an element .fr in C,,( T, X) with 0 E P,(f,) such that N,, is not open. Then ,fI is not in V, since otherwise P,,(,f,) =: {fi: and N!, = T would be open. Without loss of generality we may assume il.f, !, = I. Now let c1 ,.... r,; be linear independent elements in P,(f,) which span the linear subspace I’, :== span P,.(,f,) of 1’. Since Nf, is not open, there is a point t,, in Nf, with the property: (E) every neighborhood U of t, contains r,,(t u) F 0 for at least one K E {I ,..., k:
some point
tu such that
Now we construct a functionfas follows. In the case f,, is in Mr, , we define j’: ,f; If t, is not in MI,, then we first choose an element r in X such that
96
BROSOWSKI
j/r / -~~ I and ~I’ --f;(t,);‘,Y
AND
WI&MANN
1 -~ ‘J;(f,,)i~,t . A possiblechoice is, for instance. r
.f,(h)i
l,f;(r”)
x
in the case,f;(t,,) m=0. If J;(r,,) 0. each I’ with ,’ Y I will do. Since t,, is not in the closed set M,., . there is a compact neighborhood C; of t,, such that M,, n ZJ A= :‘. From I,, F IV’, there follows for K I..... x J’.,(t,,)
0
and
By reducing c/‘. if necessary.we can ensurethat. for all I E V. f;(t) - c.,\(t)1 -: I.
K _ 1,.... I,
Lid r
.l;ct,t
0.
There exists a continuous function pi(t) such that 0 pl(/) ’ / t T, ~,(t,,) = 1, and pi(t) = 0 for I e U. in addition, we put
I for ail
and p:%(t):
min(p,(t), max(0, p.,(t))).
We complete the definition of p2 and p3 by setting p?(t) per) -7 0 for those t which have Y-.f,(t) ~~ 0. and thus obtain a continuous function p3 with 0 zg p,(t) 3, 1 for all I E T, p,(t,,) I, and p&t) ~7.e 0 for t &IT’,,U. Now we define the functionSby .f(r) :-- (1 - p:l(f))f;(f)
1- p:,(t) . I’.
This function has the property that f(r) ,&(t) for all I in T b’, and f(t,) = r, whence ilf(t,,)ll =- t and t, E iMt . Since T\U is. by construction, an open set containing Mf, , it follows that M,, C 151~)0 EP,,(J), and finally from Lemma 8, P,(f) C V, . For each t in Tand each K = I...., k we have
CONTINUITY
OF
Hence all 1.r . . ... 13,.are in P,,(f) r,, :=
THt
METRIC
97
PROJECTlOX
and, since 0 is also in P,.(f), the element (rl
t-
is a relative interior point of P,.(f). Pv(g) where g := j - 1’” . For each t in T. we have
...
:- r,J/(k
-1 1)
Then 0 is a relative interior
with equality if and only if r,>(t) = 0 for all there follows
K
point of
= I,..., k. that is f t .\‘, . Hence
Because t, is not an interior point of IV, , there is a net (t,\ : A E 11) of points t;, in T such that (t,\) converges to t, and, for each A, c,Jt,) :. 0 at least for one K. Then there exists an index K” and a subnet (t,, : x E ‘4,) such that ~,,~(r,,) P- 0 for ah t,, with x E fl, . We may assume Kg = 1. Now we consider two cases. q
First CUSP. There is a subnet (tn : h E AJ of (tn : h E A,) such that for every A E A, there exists some functional .xA’ E L,,c~,) with xn’(r r(f,,)) it 0. By passing once more to a subnet (t,, : ,\ E L‘LJ, we can ensure that there exist signs E, t {- 1, + 11 such that the inequalities
l ,X,‘(~~,(~,,>> <: 0 and %.~A’(I’,,(I,l)) .‘, 0, hold. For each 6
:i
0, the
for
K
=
I&...,
k,
set
As := {t E T : i, g(t,,) - g(t)il < 81 is a neighborhood of t,, . Hence there exists h G A, such that t,, t A6 . Since M, C iV, and t,, 6 N, , it follows that tA $ M, . Since M, is closed, .there exists a compact neighborhood W of tA such that M, n W = G. Without loss
98
BROSOWSKI
A411
\‘v’tGklA~\
of generality. we assume CV C A,? . Now let p be a continuous function such 0 I for t F 7: p(t,,) 0 for f z 7‘ If’. and define p(t) 1. p(l)
that
,cdt) : - p(f)S(l,,) Then the function have for all f -- 7
(1
go is in C,,(7: A’). and
p(r))n(l). $.:
.p
ii. F~urthermore.
we
and for f 7’ W the equality g,(t) --- g(t). The set 7’ W is a neighborhood of hl,: Therefore we have M,,6 3 M,, , and hence 0 E P, (x6). Using Lemma 8. we obtain P, (g,s) C C:, . For each element II in the set
s
$0 .
and consequently BI,.n P,(g,) = ST. If the net (f,, : X E AI) does not satisfy the conditions of the first case, then we must consider the alternative possibility. Second cusp. There is a subnet (t, : h :- ‘1,) of (1,. : A :. I,) such that x’(rl(f,,)) 0 for all h E A, and x’ E d,,(,,) Let I,, ’ be an arbitrary functional in C,,(,,,) ,,I,A, . Then x,,’ is not in c!,~(+,) , since X is strictly convex. and r.,(t,,) is not proportional to g(t,,). Hence it follows that
On the other hand we have, for ~3’GG,,(,,,J.
and therefore
(‘ONFIUUITY
whence it follows
OF THE
blI'TRIC
99
PROJECTION
that : g(r,,) -c r,(t,+)l ;> 1
(5)
and x,‘(c,(t,)) .-. 0 for each A in A, . There are signs E, : : -I and E., , Ed ,.... E,~E {suitable subnet (r,, : h E r/l,) the inequalities
I, em I ] such that for a
and %.~,\‘(I.,(f,\)) “, 0,
for
K -=
2,
3:...,
k,
hold. For 6 ‘:, 0, the set ,4a := {t t T: ’ g(t,J - g(r);1 < 6/3, /I tll(t),l < S/3 and :I g(t,) + I’~(r)l’ --- I 1 < 6/3) is an open neighborhood of t,, . Hence there is h E A, such that r, t As. Furthermore. there exists a compact neighborhood W of t, such that Mgn Wp and WCA8, and there is a continuous function p such that 0 :’ p( ) ‘-1 1 for t E T. p(t,J = 1, and p(t) = 0 for t E T\, W. The mapping
is in C,,(7’, X). By using (5) we obtain for t in A,
.-, 1g(rJ - g(t)11 + 11Lj(t&
-t / 1 - I’ g(t,) + z.,(t,)l ~ --I (5.
For t not in A, the equality gs(t) = g(t) holds, and hence I/ g, - g I < 6. By construction, we have 11g, I! := 1. Since gs(f) = g(t) for t in the neighborhood T\ W of M, , it follows that M,,s r) M, , hence 0 is in P,(g,), and, by Lemma 8, ~l,kJ c Vl . For each element u in
100
whence B,, A P,,(g,) Thus in either case we have the following situation. The set B, is open (in V,). and contains the zero element in its boundary. Since 0 is an interior point of P, (.F) (relative to VI), it follows that P,-(g) n B,, On the other hand for each 6 ‘,. 0 there exists a g, in C&7-. X) with j g ,Y!i -:’ 6 and Pv(g6) n B,: :: . This contradicts the lower semicontinuity of P, Thus the theorem is proved. Renmks. In the caseTcompact and X the real axis, Theorem 9 specializes to a result of Blatter, Morris, and Wulbert [2]. The specialcaseof Theorem 9, when X is a pre-Hilbert space,was proved by Brosowski c’t a/. [3]. but only for locally compact spaces7‘which have additional properties. REFERENCES 1. J.
BLAI
Rlrrini.wh-
rt~.
Zur
Stetigkeit
I.t’~,.rr/d/i.\r./?pn
Imt.
van
Instwtt~.
Mrrth.
Utrir.
2. J. BIUTTER, P. D. MOKKIS, AND D. E. WULBEKT, projection. Math. Ann. 178 (1968), 12 -74. 3. B. B~osows~~. K.-H. HOFFMANN, E. SCGrI-K, metrische 4.
Projektionen
Max-Planck-Institut B. B~osowsitr,
I. Metrische fiir
K.-H.
metrischen
mengenwertigcn
Projektionen
Physik und Astrophysik HOFFMANN. E. SCti;iFFR,
metrischc Projektionen IV. Einc hinreichende metrischcn Projektion, Max-Planck-lnstitut MPI-PAE,‘Astro 20, 1969. 5. H. HASH\. “Reelle Funktionen I,” Chclsca, 6. A. J. LALIR, D. E. WULBERT, AND P. D.
SW. A. Kr.
Continuit)
of
4h11
H.
auf
lincare
Miinchen, ,\NII H. fiir
Bedingung Physik
9 ( 1964).
167-l
77.
the
WI att<.
S&r
16. 1967. 3ct-valued
metric
SfctigkeitsstitLe
Tcilr:iume
van
fiir C,,(r,
MPI-PAF. Astro 18. WFREK. StctigkeitssLitze ftir und
New York. 194s. MORRIS, C‘ontinuouz
projections. ./. FunctinnulAnalysis 3 (1969), 193 -216. 7. E. MI~HAL.I., Topologies on Spaces of subsets. Trtrm. 152-181. 8. 1. SIN,ER, Some remarks on approximative compactness. ,4/,p/.
Projchtionen.
Bairn.
.‘l/nL~i~. I&T.
H ). 1969. fiir
die Z,,-Stetigkeit dctAI1rophyi;ik Miinchen.
?elcction\ nruri1. Korl/rioitlr~
.SlW.
for
metric
71
( 195 I),
Muth.
PL/WC